CRSV Economics Computation Server

"CSRV" means "computation server". Since early summer 2009, it has eight 64-bit processors (two quad-cores) and 16 GB of RAM. Faculty and staff use this machine for computation (e.g., R, Matlab, Gauss, Stata, and compiled languages).
If you are a phd student in economcs you can use this server. Contact your supervisor about arranging an account. Notice that this server is intended to be used for research purposes,  not for serving web pages or storing your email. If you want those services, you should use googlepages, gmail, or facebook.
Support is generally informal (that is, you have to ask a colleague). If you have useful software code or instructions that you would like to make available to others, please ask for an account for this web page, so you can post your information here.
Here is an alternative source on the subject: There are a number of free SSH clients available for Mac OS, Windows, and other platforms. For help (and a list of popular clients you can use), please see the following Wiki article: http://wiki.dreamhost.com/index.php/Special:Search?fulltext\ =Search&search=ssh%20client
Once you're connected you have access to a very broad variety of standard tools common to GNU/Unix environments. Support is through your peers but the following resources may be helpful should you wish to learn more about these powerful tools:

Please do not use the server for computation until you

  1. understand the "nice" command and use it with a nice level of at least +10 for any computation jobs (+19 is standard practice for any long job).
  2. understand that you are responsible for monitoring and controlling how much memory (RAM) your calculation uses/requires
       

These ensure that you do not incapacitate the machine and thus exclude all others from the common resource. You should make use of "top" (or "htop") and understand how to tune your code so it does not use too much memory (think: setting array sizes, etc). If you're using more than a couple/few gigs of RAM, your code is likely to to run out of memory and start "swapping" (caching) like mad to the disk (look it up). That will make your code go nowhere, and the machine unusable for anyone else. You could also use the command "iotop" to see whether your job is responsible for disk caching, if you notice the RAM is used up. In general, you should closely supervise any code when you first run it, to make sure it fits within a relatively small fraction of the machine's RAM. You should also build into your code ways to monitor its progress, so you can assess whether it is making good progress and will finish in good time.
Most basic procedure for running computation tasks on a commons server:
Prefix any computation command with "nice -n+19 ", such as:
nice -n+19 matlab myprog.m
nice -n+19 xstata &
And then monitor memory use and tune your code appropriately.
 

  • For running long jobs remotely, you will find very useful the following two approaches to having your job continue after you log off or disconnect (use\ wisely! You are still responsible for what your job does when you're not supervising it): "nohup" and "screen".
  • R -- You should learn and teach R instead of Stata. Let me/us know if there are some R packages you want installed.
  • Octave -- An open source and once-again-growing Matlab alternative ("clone")
  • Python -- A modern Matlab alternative (pylab, matplotlib) is just one part of this language
  • GNU fortran, C, C++: Try "f95", "gcc", "g++"
  • Commercial software licenses on this server:
    • Matlab 7.9.0.529 (R2009b)
    • There are two licenses for Stata 11 MP (ie, with parallel processing ability). To use stata, you will have to put the following into your .bashrc file (and then relogin): [?? is this out of date?]
      export PATH=$PATH:/usr/local/opt/stata:/usr/local/stata11
      If the license is in use, you can find out who is using it, most likely, with the htop command... But see below; you may be able simply to use Stata 10 instead.
      As for various other software on the server, you can use this one through a text interface:
      stata
      or if you are using an "X server" on your local machine (this is worth the effort!) you can have the full graphical interface by running
      xstata &
      from the command line.
      If you wish stata to be updated (there are regular/frequent free updates of the program and packages), let me (cpbl) know and I can do the update for all users.
    • By magic, we also have two licenses for Stata 10 MP. Follow the instructions above but run
      stata10
      or
      xstata10
      This gives us four licenses altogether.
  • GNU/Linux
  • There is no backup whatsoever of user data on this server. Make sure to back up your own data.
Syndicate content